A Markov Decision Process model for optimal policy making in the maintenance of a single-machine single-product toolset

An aspect of great importance in Semiconductor Manufacturing Systems (SMS) is that machines are subject to unpredictable failures. Since Semiconductor Manufacturing is a highly capital intensive industry, it is crucial to optimize the usage of the resources. Performing preventive maintenance (PM), if done optimally, can reduce the risk of unpredicted failures and hence minimize the cost of outages. However, performing frequent PM results in higher cycle time and WIP accumulation at the toolset. In this paper we present a method to create optimal policies for a single-server single-product workstation using a Markov Decision Process model. The optimal policy determines whether or not to perform the PM based on the WIP level and the time since last repair. We present some numerical examples to illustrate the behavior of the optimal policy under different scenarios and compare the results with some common policies such as fixed frequency PM.